def plotDETcurve(self):
"""
This method is used to plot the DET (Detection Error Tradeoff) and
save it on the disk.
"""
# Read test scores
filename = "ubm_scores_{}.h5".format(self.NUM_GAUSSIANS)
scores_dir = os.path.join(self.BASE_DIR, "result", filename)
scores_gmm_ubm = sidekit.Scores.read(scores_dir)
# Read the key
key = sidekit.Key.read_txt(os.path.join(self.BASE_DIR, "task", "test_trials.txt"))
# Make DET plot
logging.info("Drawing DET Curve")
dp = sidekit.DetPlot(window_style='sre10', plot_title='Scores GMM-UBM')
dp.set_system_from_scores(scores_gmm_ubm, key, sys_name='GMM-UBM')
dp.create_figure()
# DET type
if self.conf['DET_curve'] == "rocch":
dp.plot_rocch_det(idx=0)
elif self.conf['DET_curve'] == "steppy":
dp.plot_steppy_det(idx=0)
else:
raise NameError("Unsupported DET-curve-plotting method!!")
dp.plot_DR30_both(idx=0) #dotted line for Doddington's Rule
prior = sidekit.logit_effective_prior(0.001, 1, 1)
dp.plot_mindcf_point(prior, idx=0) #minimum dcf point
# Save the graph
graphname = "DET_GMM_UBM_{}.png".format(self.NUM_GAUSSIANS)
dp.__figure__.savefig(os.path.join(self.BASE_DIR, "result", graphname))
def compute_eer(args):
print('Compute eer', end='')
key = sidekit.Key(os.getcwd() + '/task/dev_key.h5')
score = sidekit.Scores(os.getcwd() + '/task/dev_score.h5')
print('\rCompute eer done')
dp = sidekit.DetPlot()
prior = sidekit.logit_effective_prior(0.01, 1, 1)
dp.set_system_from_scores(score, key)
minDCF, _, __, ___, eer = sidekit.bosaris.detplot.fast_minDCF(
dp.__tar__[0], dp.__non__[0], prior, normalize=True)
with open('task/dev_result.txt', 'w') as f:
f.write('eer : {:7.2%}\n'.format(eer))
f.write('minDCF : {:7.2%}\n'.format(minDCF))
return eer, minDCF
def plotDETcurve(self):
# Read test scores
filename = "ubm_scores_{}.h5".format(self.NUM_GUASSIANS)
scores_dir = os.path.join(self.BASE_DIR, "result", filename)
scores_gmm_ubm = sidekit.Scores.read(scores_dir)
# Read the key
key = sidekit.Key.read_txt(os.path.join(self.BASE_DIR, "task", "test_trials.txt"))
# Make DET plot
logging.info("Drawing DET Curve")
dp = sidekit.DetPlot(window_style='sre10', plot_title='Scores GMM-UBM')
dp.set_system_from_scores(scores_gmm_ubm, key, sys_name='GMM-UBM')
dp.create_figure()
dp.plot_rocch_det(0)
dp.plot_DR30_both(idx=0)
prior = sidekit.logit_effective_prior(0.01, 10, 1)
dp.plot_mindcf_point(prior, idx=0)
graphname = "DET_GMM_UBM_{}.png".format(self.NUM_GUASSIANS)
dp.__figure__.savefig(os.path.join(self.BASE_DIR, "result", graphname))
# Compute all trials and save scores in HDF5 format
print('Compute trial scores')
scores_gmm_ubm = sidekit.gmm_scoring(ubm,
enroll_sv,
test_ndx,
features_server,
num_thread=nbThread)
scores_gmm_ubm.write('scores/scores_gmm-ubm_rsr2015_male.h5')
# Plot DET curve and compute minDCF and EER
print('Plot the DET curve')
# Set the prior following NIST-SRE 2008 settings
prior = sidekit.logit_effective_prior(0.01, 10, 1)
# Initialize the DET plot to 2008 settings
dp = sidekit.DetPlot(window_style='sre10', plot_title='GMM-UBM_RSR2015_male')
dp.set_system_from_scores(scores_gmm_ubm, key, sys_name='GMM-UBM')
dp.create_figure()
dp.plot_rocch_det(0)
dp.plot_DR30_both(idx=0)
dp.plot_mindcf_point(prior, idx=0)
print('Plot DET curves')
prior = sidekit.logit_effective_prior(0.001, 1, 1)
minDCF, Pmiss, Pfa, prbep, eer = sidekit.bosaris.detplot.fast_minDCF(
dp.__tar__[0], dp.__non__[0], prior, normalize=True)
print("UBM-GMM 128g, minDCF = {}, eer = {}".format(minDCF, eer))
itNb=(it, 0, 0),
minDiv=True,
ubm=None,
batch_size=1000,
numThread=nbThread)
scores_plda_efr1 = sidekit.iv_scoring.PLDA_scoring(
enroll_iv_sn1, test_iv_sn1, test_ndx, mean1, F1, G1, Sigma1)
scores_plda_efr1.save(
'scores/scores_plda_rank_400_it10_efr1_sre10_coreX-coreX_m.h5')
if plot:
print('Plot the DET curve')
# Set the prior following NIST-SRE 2010 settings
prior = sidekit.effective_prior(0.001, 1, 1)
# Initialize the DET plot to 2010 settings
dp = sidekit.DetPlot(windowStyle='sre10',
plotTitle='I-Vectors SRE 2010-ext male, cond 5')
dp.set_system_from_scores(scores_cos, keysX[4], sys_name='Cosine')
dp.set_system_from_scores(scores_cos_wccn,
keysX[4],
sys_name='Cosine WCCN')
dp.set_system_from_scores(scores_cos_lda, keysX[4], sys_name='Cosine LDA')
dp.set_system_from_scores(scores_cos_lda_wcnn,
keysX[4],
sys_name='Cosine WCCN LDA')
dp.set_system_from_scores(scores_mah_efr1,
keysX[4],
sys_name='Mahalanobis EFR')
dp.set_system_from_scores(scores_2cov, keysX[4], sys_name='2 Covariance')
nap_iv_sn1 = copy.deepcopy(nap_iv)
enroll_iv_sn1 = copy.deepcopy(enroll_iv)
test_iv_sn1 = copy.deepcopy(test_iv)
nap_iv_sn1.spectral_norm_stat1(meanSN[:1], CovSN[:1])
enroll_iv_sn1.spectral_norm_stat1(meanSN[:1], CovSN[:1])
test_iv_sn1.spectral_norm_stat1(meanSN[:1], CovSN[:1])
print('Run PLDA rank = {}, {} iterations with Spherical Norm'.format(400, 10))
mean1, F1, G1, H1, Sigma1 = nap_iv_sn1.factor_analysis(400, rank_G=0, rank_H=None,
re_estimate_residual=True,
itNb=(10,0,0), minDiv=True, ubm=None,
batch_size=1000, numThread=nbThread)
scores_plda_sn1 = sidekit.iv_scoring.PLDA_scoring(enroll_iv_sn1, test_iv_sn1, test_ndx, mean1, F1, G1, Sigma1)
scores_plda_sn1.save('scores/scores_plda_eigh_rank_{}_it{}_lapack_sn1_sre10_coreX-coreX_m_mfcc.h5'.format(400, 10))
print('Plot PLDA scoring DET curves')
prior = sidekit.effective_prior(0.001, 1, 1)
# Initialize the DET plot to 2008 settings
dp = sidekit.DetPlot(windowStyle='sre10', plotTitle='I-Vectors SRE 2010 male')
dp.set_system_from_scores(scores_plda_sn1, keysX[4], sys_name='Cond 5')
dp.set_system_from_scores(scores_2cov, keys[4], sys_name='PLDA Spherical Norm')
dp.create_figure()
dp.plot_rocch_det(0)
dp.plot_DR30_both(idx=0)
dp.plot_mindcf_point(prior, idx=0)